Tencent / ncnn

ncnn is a high-performance neural network inference framework optimized for the mobile platform
Other
20.15k stars 4.14k forks source link

实测树莓派3b的benchmark时间过长 #795

Closed ljpadam closed 5 years ago

ljpadam commented 5 years ago

mobilenet和mobilenetssd在树莓派3b上运行时间较reamde中的时间分别多了30ms和80ms,但是其他网络的运行时间基本正常。编译步骤和运行方式完全和指南中的一致,请问可能是什么原因造成mobilenet的运行时间过长?

下面是在树莓派3b上运行benchmark后的结果, pi@raspberrypi:~/code/ncnn2/ncnn/build/benchmark $ ./benchncnn 8 4 0 loop_count = 8 num_threads = 4 powersave = 0 gpu_device = -1 squeezenet min = 102.85 max = 114.07 avg = 105.64 mobilenet min = 188.02 max = 196.52 avg = 189.66 mobilenet_v2 min = 199.89 max = 219.67 avg = 204.00 shufflenet min = 72.04 max = 81.76 avg = 73.70 mnasnet min = 132.11 max = 144.29 avg = 134.11 proxylessnasnet min = 162.51 max = 174.80 avg = 167.04 googlenet min = 332.20 max = 376.71 avg = 349.60 resnet18 min = 501.01 max = 543.74 avg = 523.46 alexnet min = 414.20 max = 478.88 avg = 452.09 vgg16 min = 2453.36 max = 2549.59 avg = 2501.66 squeezenet-ssd min = 340.02 max = 354.93 avg = 346.49 mobilenet-ssd min = 344.44 max = 377.71 avg = 356.27 mobilenet-yolo min = 901.09 max = 950.96 avg = 927.55 mobilenet-yolov3 min = 777.89 max = 806.25 avg = 790.33

下面是benchmark中readme给出的结果 pi@raspberrypi:~ $ ./benchncnn 8 4 0 loop_count = 8 num_threads = 4 powersave = 0 squeezenet min = 108.66 max = 109.24 avg = 108.96 mobilenet min = 151.78 max = 152.92 avg = 152.31 mobilenet_v2 min = 193.14 max = 195.56 avg = 194.50 shufflenet min = 91.41 max = 92.19 avg = 91.75 googlenet min = 302.02 max = 304.08 avg = 303.24 resnet18 min = 411.93 max = 423.14 avg = 416.54 alexnet min = 275.54 max = 276.50 avg = 276.13 vgg16 min = 1845.36 max = 1925.95 avg = 1902.28 squeezenet-ssd min = 313.86 max = 317.35 avg = 315.28 mobilenet-ssd min = 262.91 max = 264.92 avg = 263.85 mobilenet-yolo min = 638.73 max = 641.27 avg = 639.87

nihui commented 5 years ago

benchmark readme 是64位系统上跑的,你的是32位系统?

ljpadam commented 5 years ago

感谢回复,我是在树莓派官方的32位系统上运行的,请问你用的64位系统是从哪里下载的呢?

nihui commented 5 years ago

这个你自己找下吧,比如 https://wiki.ubuntu.com/ARM/RaspberryPi#arm64 ...

JoeyZhu commented 5 years ago

benchmark readme 是64位系统上跑的,你的是32位系统?

除了64位,使用别的编译器会更快吗?快多少?